[bercedev]

Reference Issues/PRs

Fixes #2472.

What does this implement/fix? Explain your changes.

MultiNormalizer had no fit_transform of its own, so it used sklearn's TransformerMixin.fit_transform, i.e. self.fit(y).transform(y). That breaks in two ways:

• fit reads one column per target (y[:, idx]), but for ndarray and Tensor input transform reads one row per target (y[idx]), and only transposes a DataFrame itself. So fit_transform on an array or tensor scaled the first samples instead of the targets, and returned arrays with as many elements as there are targets.
• X never reached transform, so with a GroupNormalizer among the normalizers the call failed with either target_scale or X has to be passed.

This adds MultiNormalizer.fit_transform(y, X=None, return_norm=False). It fits on y as fit expects, transposes an array or tensor for transform (a DataFrame is left as it is), and passes X and return_norm on. The result is the same as the explicit fit(y).transform(y.T) control in the issue. Nothing in the library calls MultiNormalizer.fit_transform (TimeSeriesDataSet uses transform), so existing code paths don't change.

What should a reviewer concentrate their feedback on?

• Whether fit_transform should keep the existing transform layout (one row per target for arrays/tensors), as done here, rather than changing transform.

Did you add any tests for the change?

Yes, in tests/test_data/test_encoders.py:

• test_MultiNormalizer_fit_transform[numpy|torch|pandas]: every target keeps all 4 samples and equals a TorchNormalizer fitted on that target alone. numpy and torch fail on main, pandas already passed.
• test_MultiNormalizer_fit_transform_passes_groups: with a GroupNormalizer, the first target equals GroupNormalizer.fit_transform(data["a"], data). Fails on main with the assertion above.

tests/test_data/test_encoders.py: 62 passed (Python 3.12, CPU torch 2.14.1). ruff 0.6.9 (the pre-commit version) check and format are clean.

Any other comments?

Note: I used Claude Code while working on this.

PR checklist

• The PR title starts with either [ENH], [MNT], [DOC], or [BUG].
• Added/modified tests
• Used pre-commit hooks when committing (I ran the ruff hooks' version by hand, not pre-commit run)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (main@ca70dc8). Learn more about missing BASE report.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #2473   +/-   ##
=======================================
  Coverage        ?   88.82%           
=======================================
  Files           ?      218           
  Lines           ?    11675           
  Branches        ?        0           
=======================================
  Hits            ?    10370           
  Misses          ?     1305           
  Partials        ?        0           

Flag	Coverage Δ
cpu	88.82% <100.00%> (?)
pytest	88.82% <100.00%> (?)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.